Job files

An example job

yaml

# A job, written as a short list of needs.
job: fine-tune-clinic-notes
model: open-model            # any open model
gpu: h100 x8                 # eight NVIDIA H100s
memory: 640gb
policy: patient-data         # the rules it follows
data_location: bangladesh    # stays in the country
max_price_per_hour: auto     # set by the policy
deadline: friday 18:00
on_change: move              # re-place on any change

Fields

Field	What it does	Value in the example
`job`	The name of the job	`fine-tune-clinic-notes`
`model`	The app or model to run	`open-model`: any open model
`gpu`	The kind and number of GPUs, or leave it to the Cloud Router	`h100 x8`
`memory`	The memory the job needs	`640gb`
`policy`	The policy the job follows	`patient-data`
`data_location`	Where the job's data must stay	`bangladesh`
`max_price_per_hour`	The most the job may cost per hour	`auto`: set by the policy
`deadline`	When the job must finish	`friday 18:00`
`on_change`	What happens when price, capacity or location changes	`move`: re-place the job

Hardware you can name

NVIDIA H100, H200 and B200, AMD MI300X, and CPUs of every size. A job can name the hardware it needs or leave the choice to the Cloud Router, which checks what is free on each cloud before every placement.

Pin a job to a cloud

Pin a job to one cloud or one region and the Cloud Router keeps it there. You will rarely need to: with a policy and a location rule, the Router keeps every job where it is allowed to be, at the best price it can find.

How the job is placed

The Cloud Router filters by location and policy, checks live capacity, ranks by price, places the job and keeps watching. See The Cloud Router.

#An example job

#Fields

#Hardware you can name

#Pin a job to a cloud

#How the job is placed

An example job

Fields

Hardware you can name

Pin a job to a cloud

How the job is placed